Download as pdf or txt
Download as pdf or txt
You are on page 1of 14

This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TWC.2019.2954318, IEEE
Transactions on Wireless Communications
1

A Channel Model for Indoor Time-of-Arrival


Ranging
Zoran Latinović and Howard Huang, Fellow, IEEE

Abstract—We executed an indoor measurement campaign to improving the scalability compared to systems with pairwise
statistically characterize the bias in time of arrival (ToA) esti- ranging. For a system using outdoor bases, the bases could be
mates incurred by multipath in the 5.5 GHz industrial, scientific synchronized with satellite navigation receivers, and because
and medical band by transmitting 10 and 100 MHz bandwidth
waveforms. Measurements in three buildings with different pro- their signals have ubiquitous coverage, one could localize
pagation conditions are performed to analyze ToA statistics in both outdoor and indoor devices. Performance for indoor
a wide range of indoor environments, from partitioned offices devices could be improved significantly if nearby indoor LTE
to an open cafeteria area. We demonstrate that WINNER II base stations also transmit PRSs. Because satellite navigation
and similar channel models designed for communications are signals typically do not have indoor coverage, synchronization
not suitable for localization applications as they do not provide
absolute delays. Furthermore, such models assume constant root for the indoor infrastructure would need to be achieved through
mean square delay spread over distance, which is a significant another means, e.g., a wired [4] or wireless [5] technique. In
contributor to the ToA errors for lower signal bandwidths. addition to the scalability advantage, the LTE solution has a
Other physical aspects such as specular components in line- significant cost advantage due to economies of scale, compared
of-sight (LOS) and random bias in non-LOS scenario are also to a proprietary sensor network solution, for example based on
not considered in those channel models. An extension of the
WINNER II model that overcomes those flaws and provides ultrawideband (UWB).
more realistic ranging performance is proposed. We validate the In OTDOA localization systems, the performance is de-
extended model by demonstrating a good match between the graded due to synchronization error, inaccurate base location
statistics of its simulated output and our measurements.
estimates, and ToA estimation error [6]. As a result of unre-
Index Terms—Indoor localization, time of arrival, ranging, solvable multipath and non line-of-sight (NLOS) channels, the
channel modeling for localization, ToA bias statistics. estimated ToA could be greater than the ToA corresponding
to a single-path, line-of-sight (LOS) channel. This difference,
I. I NTRODUCTION which we refer to as bias, could be several hundred na-
noseconds, for example if the direct LOS is blocked, and
scalable, low-cost solution for accurate indoor localiza-
A tion will be a key enabling technology for smartphone
and Internet of Things applications. While there are many
the receiver measures the ToA of a reflected NLOS path. A
positive bias results in a positive range error which in turn
causes errors in the device location estimate.
classes of localization techniques, for example based on angle
of arrival or signal strength measurements, time of arrival While channel models for sub-6GHz bands given by the
(ToA) based localization achieves more accurate performance LTE standard [7] or WINNER II [8] are suitable for com-
for higher signal bandwidths [1]. Much of the literature on ind- munications, we claim that they are not suitable for ToA-
oor localization based on ToA measurements assumes that the based localization because they fail to capture a number
location estimate of a device is derived from multiple pairwise of relevant characteristics. As a result of this shortcoming,
ranging estimates to anchor nodes with known location [2]. the measured bias can be unrealistically low, and the lo-
This solution would not be suitable for a system with many cation estimate can be unreasonably accurate. Experimental
devices that each requires a rapid update rate on position. measurements are more expensive than models but provide
A more scalable solution would be one based on the Long- a more realistic characterization of performance. A recent
Term Evolution (LTE) cellular standard known as observed paper describes a measurement campaign using UWB to
time difference of arrival (OTDOA) [3]. The principle of LTE characterize ToA based ranging in underground mines [9].
OTDOA is similar to satellite navigation systems such as GPS Experimental measurements in an indoor office are used to
or Gallileo, except that synchronized cellular base stations, evaluate various localization algorithms in [10]. However the
instead of satellites, transmit positioning reference signals underlying ToA ranging estimates were provided by hardware
(PRSs) from which timing is measured. Systems with synchro- not under the control of the authors, and there were significant
nized infrastructure enable simultaneous ranging estimates, negative ranging errors that could not be explained. In [6], the
performance of an OTDOA system was evaluated using ToA
Z. Latinović is with Nokia Mobile Networks, Murray Hill, NJ 07974 USA measurements from a live LTE network, but again the authors
(e-mail: zoran.latinovic@nokia.com) and New York University Tandon School had no control of the ToA estimation algorithm. Authors
of Engineering, Brooklyn, NY 11201 USA.
H. Huang is with Nokia Bell Labs, Murray Hill, NJ 07974 USA (e-mail: in [11], [12] obtained NLOS bias statistics from measurements
howard.huang@nokia-bell-labs.com). at 5.2 GHz for outdoor-to-indoor channel. The ToA ranging
0000–0000/00$00.00
c 2019 IEEE

1536-1276 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TWC.2019.2954318, IEEE
Transactions on Wireless Communications
2

error has been modeled from UWB measurements in [13]. B. Layouts of the Measurement Areas
Channel measurements in 5.5 GHz band has been performed
in a closed indoor space to collect ToA bias statistics for The indoor measurement campaign was conducted at the
10 and 100 MHz bandwidth waveforms in [14]. This paper Nokia Murray Hill campus in New Jersey. Measurements
is continuation of that work and its main contributions are are performed in three buildings with wide range of indoor
summarized below. propagation environments, from partitioned offices (closed
space) over typical cubical space to very open cafeteria mall
1) We conduct a measurement campaign over three dif- area, Fig. 1 top, center and bottom respectively.
ferent indoor environments and characterize the ToA
bias error statistics. Compared to real measurements,
WINNER II [8] and similar communication channel
✶ ✷ ✸ ✹
models widely used today result in unrealistically small ✶

ToA ranging error leading to optimistic localization ✶✦ ✷✦ ✸✦ ✶✦✦ ✹✦


✶ ✦✦❛ ✺✦ ✻
❝✁✂✂✄☎✁✂ ✶
performance.
✸✦✦ ✾
2) We identify a number of deficiencies of existing commu- ✷✦✦❛
nications channel models, including a lack of absolute ✻✦
path delays and a constant rms delay spread as a function
of distance. Other physical aspects of the wireless ✺
✶ ✦✦❜ ✾✦✦
✽ ✷✦✦❜
❝✁✂✂✄☎✁✂ ✷
channel such as specular components or reflections in ✼✦ ✶ ✦✦❝

LOS and random bias in NLOS scenario are also not ✽✦

considered in those channel models.


3) We propose an extension of the WINNER II model (a) Closed space
with additional parameters to account for the identified
deficiencies. The parameter values are chosen to match
mean and rms delay spread statistics between the model
✹✦✦
and our measurements. For two ToA estimation met-
hods, we show a good match between the ToA statistics
of the extended model and measurements. ✸✦✦❛
✸✦✦❜
✸✦✦❝ ✶✦❛ ❝ ✁✁✂✄ ✁ ✹
✷✦❛ ✻✦✦✦

In Section II, we describe a custom channel sounding
✷✦❜
system with synchronized transceivers, capable of providing
☎✆ ✶✦❜ ✷✦❝
☎✆ ✻ ❝ ✁✁✂✄ ✁ ✶ ✝✝
high-resolution baseband samples. From these samples, we ✝✝
✞✟
✞✟

✆ ✝

perform offline ToA estimation and generate statistics of the ✝
✠ ✶ ✷

ToA bias error. Measurement campaign in three different ✹


✺✦✦✦
✶✦❝
indoor environment types is described and mean and rms delay ✸
spread statistics calculated from the measurements. Extended
WINNER II model suitable for ranging is proposed and fitting
(b) Semi-open space
procedure for finding its parameters is explained in Section III.
The ToA error comparison between the simulated extended
✶✸
model and measurements is given in Section IV. Comparison ✶✹✦✦ ✶✻✦✦
✽✦
between the WINNER II and extended model is also included
in the same section. Finally, some concluding remarks are ✺
✼✦
✶✸✦✦ ✸ ✻
made in Section V. ✽ ✺✦✦
✻✦✦ ✶✷
✶✺✦✦

✸✦ ✶✷✦

✶✹ ✼
✶✻
II. I NDOOR T OA M EASUREMENT C AMPAIGN ✶ ✶✦
✷✦
✷ ✶✶✦ ✶ ✶✼✦✦ ✹✦✦

✾✦
A. Channel Sounder Architecture ✶✶

✶ ✦
The channel sounder system transmits/receives signals of ✶✼
✶✺

bandwidth B ∈ {10, 100} MHz at 5.54 GHz carrier frequency.


(c) Open space
Starts of transmissions and captures are synchronized by
common 1 pulse-per-second (PPS) trigger, while frequencies Fig. 1. Closed space (1a) layout with NLOS measurements. LOS measure-
ments (not shown) are taken in the corridors 1 and 2. Semi-open space (1b)
and phases of internal oscillators are synchronized by a layout with NLOS measurements. LOS measurements (not shown) are taken
single 10 MHz generator. The sounder employs a maximum- in corridors 1, 2, 3 and 4. Open space (1c) layout with NLOS (clusters) and
length binary pseudo-noise sequence (m-sequence) of length LOS (arcs) measurements. Pairs of numbers indicate Tx (red dots) and Rx
(blue crosses) arcs and clusters locations taken at distances of 10, 20 or 30
M = 8191 chips with root-raised cosine pulse shaping. Sliding m denoted as n-n0 , n-n00 and n-n000 , respectively. Antenna heights are 2 m
correlator is performed offline by correlating received captures except for Bldg 7 where heights are 1.5 m and 2 m for Tx and Rx antennas,
with the transmitted m-sequence. More details about the used respectively.
channel sounder are provided in the Section II-A of [14].

1536-1276 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TWC.2019.2954318, IEEE
Transactions on Wireless Communications
3

TABLE I
S UMMARY OF M EASUREMENTS IN C LOSED , S EMI - OPEN AND O PEN S PACES

Num. of locations Num. of measurements


Scenario Locations
10 m 20 m 30 m 10 m 20 m 30 m
Closed LOS corridor 1 18 20 858 655
space corridor 2 20 17 637 551
NLOS clusters n-n0 60 1883
clusters n-n00 56 1849
Semi- LOS corridor 1 14 19 46 420 570 1380
open corridor 2 14 420
space corridor 3 16 8 480 240
corridor 4 14 420
NLOS clusters n-n0 42 1260
clusters n-n00 42 1260
clusters n-n000 44 1320
Open LOS arcs n-n0 82 2460
space arcs n-n00 69 2070
NLOS clusters n-n0 20 600
clusters n-n00 23 690

Closed space, roughly 30×18 m2 area, is partitioned into C. Mean and RMS Delay Spread
private offices and storage rooms with hallway walls that The mean excess delay τm is defined as the first central
are mostly covered with metal sheets. Semi-open space of moment
approximately 85×30 m2 is partitioned into 6 sections, with
PL 2
τl |h(τl )|
thick structural concrete walls placed every 30 m. Each section τm = Pl=1L 2
, (1)
is a large room of size 13×28 m2 turned into cubical spaces l=1 |h(τl )|

by light panels made out of vinyl, polyester and fabric with and rms delay spread τrms as the square root of the second
steel and aluminum framing. Open space is close to the main central moment of the power delay profile
entrance and much more open area compared to two other v
u PL 2 2
buildings. This large area is composed of the cafeteria and u (τl − τm ) |h(τl )|
τrms = t l=1PL , (2)
dining room that together measures 55×14 m2 of open space. 2
l=1 |h(τl )|

where h(τl ) is the time domain channel impulse response


obtained from the measurement data.
Unlike τrms , obtaining statistics of the τm requires not
We rather focus on obtaining ToA error statistics for fewer only well synchronized Tx and Rx devices (concurrent start
distances (10, 20 or 30 m) between the transmitter (Tx) and of the Tx transmission and Rx capture or time mismatch
receiver (Rx) than having random distances typically done in with some fixed and known delay that can be subtracted in
channel measurements. Taking measurements at fewer discrete the calibration process) but also accurate Euclidean distance
distances can better reveal ToA error dependence on the Tx/Rx between the Tx and Rx antennas. Information about the
separation. Knowing the exact separation between the Tx and distance between Tx and Rx is often omitted in communication
Rx antennas is crucial for determining the ToA ranging error, channel measurement campaigns since their main interest is
which in the LOS conditions is measured by using a high- finding the τrms .
precision laser distance finder. It is more challenging for the The τm and τrms delay spread are calculated from each
NLOS case since there is no visible line of sight and we had to measured channel impulse response ĥ(τ ) obtained from the
measurep sides and calculate Euclidean distance indirectly by correlator profile for 100 MHz signal. The noise threshold
d = (xT x − xRx )2 + (yT x − yRx )2 + (zT x − zRx )2 , where is estimated from samples prior to the direct path arrival.
(xT x , yT x , zT x ) and (xRx , yRx , zRx ) are coordinates of the Tx Correlator samples below the noise threshold are replaced by
and Rx antennas in a three-dimensional coordinate system. zeros and the τm and τrms are calculated only with remaining
Since our intention is to take lots of measurement at only few samples, whose cdfs are shown in Fig. 2. Mean and standard
distances, lengths between two locations are measured and deviation of the measured τm and τrms are listed in Table II.
then clusters like in Fig. 1 (red dots and blue crosses on 2D
grid) are formed along the grid around those two reference
points to reduce the tedious work of measuring distances for D. Sliding Correlator ToA Estimator with threshold β
every pair of locations. Both carts had to be moved to the mat- A conventional ToA estimator that selects the time corre-
ching pair of grid points keeping the mutual distance constant. sponding to the peak of correlator absolute value |C(τ )|2 is op-
For each location in clusters approximately 30 measurements timal in the AWGN channel. This estimator could miss earlier
on average are taken for each signal bandwidth (Table I). We weaker paths, especially for signals with higher bandwidths
were walking randomly in the measurement area to simulate in NLOS environments. We use a robust and computationally
real life environment and obtain different channel realizations. simple peak search algorithm in [15], which estimates the ToA

1536-1276 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TWC.2019.2954318, IEEE
Transactions on Wireless Communications
4

(a) Closed space, mean delay spread (b) Closed space, rms delay spread

(c) Semi-open space, mean delay spread (d) Semi-open space, rms delay spread

(e) Open space, mean delay spread (f) Open space, rms delay spread
Fig. 2. Empirical mean (left) and rms (right) delay spread cdfs for the closed, semi-open and open spaces with black dotted lines representing cdfs drawn
from fitted log-normal distribution. Means and standard deviations of τm and τrms are given in Table II. Thick black solid and dashed lines represent the
WINNER II A1 channel model for 10 m and closed space (LOS and NLOS curves for 20 m are omitted since they are identical to those of 10 m).

1536-1276 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TWC.2019.2954318, IEEE
Transactions on Wireless Communications
5

TABLE II
M EASURED MEAN (µ) AND STANDARD DEVIATION (σ) PARAMETERS OF THE τm AND τrms DELAY SPREAD , T OA ERROR τ̂B AND CORRELATION
COEFFICIENTS ρ(τ̂B , τm ) AND ρ(τ̂B , τrms ) FOR SIGNAL BANDWIDTHS B = 10 AND B = 100.

τm (ns) τrms (ns) τ̂10 (ns) τ̂100 (ns) ρ(τ̂B , τm ) ρ(τ̂B , τrms )
Scenario
µ σ µ σ µ σ µ σ B=10 B=100 B=10 B=100
Closed LOS 10 m 59 20 76 18 39 40 8 10
0.37 0.2 0.16 0.01
space 20 m 46 24 68 30 27 31 7 10
NLOS 10 m 73 15 57 11 57 40 25 13
0.57 0.83 0.32 0.36
20 m 121 33 72 11 100 53 60 35
Semi- LOS 10 m 35 14 45 14 19 28 3 5
open 20 m 25 13 37 11 11 23 3 6 0.46 0.18 0.32 0.04
space 30 m 20 14 38 15 10 21 2 2
NLOS 10 m 46 18 47 11 34 42 8 8
20 m 53 23 47 10 44 46 9 10 0.56 0.56 0.42 0.38
30 m 101 24 78 13 80 53 22 16
Open LOS 10 m 12 5 26 8 4 17 1 2
0.39 −0.17 0.18 −0.18
space 20 m 17 7 32 11 13 22 3 3
NLOS 10 m 48 21 32 10 45 29 26 18
0.49 0.51 0.27 −0.18
20 m 49 13 39 13 36 34 22 14

as the time corresponding to the first local peak which exceeds model does not provide absolute delays; 2) ToA error cdfs
some fraction β of the overall maximum peak: for 10 and 20 m are identical due to the fact that modeled
 h i channel delay spread is independent of the Tx/Rx distance.
τ̂ = arg min |C(τ )|2 ≥ β 2 max |C(α)|2 : Separation between Tx and Rx has influence only on the path-
τ α
 loss, which does not affect ToA estimation significantly if the
2 2
|C(τ + a)| < |C(τ )| , ∃ > 0, |a| < , a 6= 0 . (3) signal’s SNR is sufficiently high [3]; 3) WINNER II model
results in unrealistically small ToA error compared to the real
A simple data dependent method for determining threshold β measurements.
is proposed. Starting from 1, βi,j,k for environment i, distance
Following observations can be made from the cdf plots in
j and propagation k is being lowered until the variance
Fig. 3 and data in Table II: 1) ToA error is larger for the 10
σ 2 (τ̂ (βi,j,k )) reaches minimum. This process is repeated for
MHz signal compared to the 100 MHz waveform since larger
all i, j and k and the maximum β̂ among all βi,j,k is selected
signal bandwidth can better resolve multipath components; 2)
as the single unique threshold. The single threshold β̂ is not
Metal walls in the closed space create significant ToA bias
optimal for all environments and distances but it is chosen for
error, up to a couple hundred nanoseconds and a hundred na-
the sake of generality. Threshold β̂ obtained by this algorithm
noseconds, for 10 MHz and 100 MHz waveforms respectively;
is 0.5 for 100 MHz and 0.7 for 10 MHz signal bandwidths.
3) All LOS and most NLOS cdfs show not only positive but
also negative errors (with higher probability for 10 MHz) or
E. ToA Bias Error CDFs estimated ToA delays that are smaller than the direct path.
ToA estimates τ̂ are calculated according to (3) and their This type of error could occur, for instance, in two-ray model
cdfs are shown in Fig. 3. The ToA cdfs obtained from if the reflected path arrives in a counter phase to the direct
WINNER II channel model simulations are also shown for path, in which case the superposition of unresolvable paths
the closed space in Figs. 3a and 3b. The A1 indoor office creates a peak with a delay smaller than the direct path like
scenario is considered with the path-loss options set to match in [16] and [17, Fig. 5]; 4) It is surprising that 10 MHz LOS
the measurement environment. NLOS locations for the closed ToA error decreases with distance for the closed and semi-
space are classified based on the propagation conditions as open space (Fig. 3a and 3c) but increases with distance for the
RR light with two walls (cluster pair 7-7’ in Fig. 1 top) and open space (Fig. 3e). Similar trend is noticeable for the τm
CR heavy from one to four walls between the Tx and Rx for and less for the τrms in Fig. 2 and Table II. Our assumption is
all other locations1 . Transmit waveforms are convolved with that for smaller distances delays of reflections are further from
independent channel realizations, noise with power similar to the direct path, which increases the τm . In contrast, delays of
those of measurements is added to the received signal, which reflections are closer to the LOS for larger distances and the
is then processed regularly. τm is lower compared to the small distance case. There are
There are several shortcomings of utilizing WINNER II no strong reflections for the open space resulting in expected
model in localization applications: 1) exact distance between τm increase over distance. The common trend between the
the Tx and Rx for NLOS channels is unknown, i.e., the ToA error and delay spread is supported by moderate positive
correlations ρ(τ̂ , τm ) and ρ(τ̂ , τrms ) in Table II; 5) ToA error
1 In the A1 scenario [8, Table 4-4], modeled path-loss in NLOS environment for 100 MHz waveform in LOS channel is similar for all
depends not only on the carrier frequency and distance between the Tx and distances since larger signal bandwidth can better isolate first
Rx but also on their positions that can be classified either as CR (corridor-
to-room) or as RR (room-to-room), number of walls between the Tx and Rx arriving path; 6) NLOS ToA errors increase with increasing
and their structures (heavy or light). distance for all buildings and both signal bandwidths; 7) Based

1536-1276 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TWC.2019.2954318, IEEE
Transactions on Wireless Communications
6

✁✵✎ ✁✵✌

✁✵✍ ✁✵✽

✁✵✌ ✁✵☞

✁✵☞ ✁✵✻
✑ ✎
✏ ✁✵✂ ✍ ✁✵☛
❈ ❈
✁✵☛ ✁✵✹

✁✵✸ ▲✒✓ ✔✕✖✗✘ ✁✵✡ ▲✏✑ ✒✓✔✕✖


◆▲✒✓ ✔✕✖✗✘ ◆▲✏✑ ✒✓✔✕✖
✁✵✷ ▲✒✓ ✔✙✖✗✘ ✁✵ ▲✏✑ ✒✗✔✕✖
◆▲✒✓ ✔✙✖✗✘ ◆▲✏✑ ✒✗✔✕✖
✁✵ ▲✒✓ ✔✚✛◆◆✜✢ ✛✛✘ ✁✵✶ ▲✏✑ ✒✘✙◆◆✚✛ ✙✙✖
◆▲✒✓ ✔✚✛◆◆✜✢ ✛✛✘ ◆▲✏✑ ✒✘✙◆◆✚✛ ✙✙✖
✁ ✁
✲ ✁✁ ✲✂✁ ✁ ✂✁ ✁✁ ✂✁ ✷✁✁ ✷✂✁ ✸✁✁ ✲ ✁ ✁ ✁ ✹✁ ✻✁ ✽✁ ✶✁✁ ✶ ✁ ✶✹✁
❉✄☎✆✝ ✞✟✠✡ ❉✂✄☎✆ ✝✞✟✠
(a) Closed space, 10 MHz (b) Closed space, 100 MHz

✁✵✏ ✁✵✌

✁✵✎ ✁✵✽

✁✵✍ ✁✵☞

✁✵✌ ✁✵✻
✒ ✎
✑ ✁✵✂ ✍ ✁✵☛
❈ ❈
✁✵☞ ✁✵✹

✁✵☛ ▲✓✔ ✕✖✗✘✙ ✁✵✡ ▲✏✑ ✒✓✔✕✖


◆▲✓✔ ✕✖✗✘✙ ◆▲✏✑ ✒✓✔✕✖
✁✵✷ ▲✓✔ ✕✚✗✘✙ ✁✵ ▲✏✑ ✒✗✔✕✖
◆▲✓✔ ✕✚✗✘✙ ◆▲✏✑ ✒✗✔✕✖
✁✵ ▲✓✔ ✕✛✗✘✙ ✁✵✶ ▲✏✑ ✒✘✔✕✖
◆▲✓✔ ✕✛✗✘✙ ◆▲✏✑ ✒✘✔✕✖
✁ ✁
✲ ✁✁ ✲✂✁ ✁ ✂✁ ✁✁ ✂✁ ✷✁✁ ✷✂✁ ✲ ✁ ✁ ✁ ✹✁ ✻✁ ✽✁ ✶✁✁
❉✄☎✆✝ ✞✟✠✡ ❉✂✄☎✆ ✝✞✟✠
(c) Semi-open space, 10 MHz (d) Semi-open space, 100 MHz

✁✵✏ ✁✵✌

✁✵✎ ✁✵✽

✁✵✍ ✁✵☞

✁✵✌ ✁✵✻
✒ ✎
✑ ✁✵✂ ✍ ✁✵☛
❈ ❈
✁✵☞ ✁✵✹

✁✵☛ ✁✵✡

✁✵✷ ▲✓✔ ✕✖✗✘✙ ✁✵ ▲✏✑ ✒✓✔✕✖


◆▲✓✔ ✕✖✗✘✙ ◆▲✏✑ ✒✓✔✕✖
✁✵ ▲✓✔ ✕✚✗✘✙ ✁✵✶ ▲✏✑ ✒✗✔✕✖
◆▲✓✔ ✕✚✗✘✙ ◆▲✏✑ ✒✗✔✕✖
✁ ✁
✲ ✁✁ ✲✂✁ ✁ ✂✁ ✁✁ ✂✁ ✷✁✁ ✲ ✁ ✁ ✁ ✹✁ ✻✁ ✽✁ ✶✁✁ ✶ ✁ ✶✹✁
❉✄☎✆✝ ✞✟✠✡ ❉✂✄☎✆ ✝✞✟✠
(e) Open space, 10 MHz (f) Open space, 100 MHz
Fig. 3. ToA error cdfs τ̂10 (left side plots) and τ̂100 (right side plots) in closed, semi-open and open spaces, and distances of 10, 20 and 30 m (the semi-open
space only). The WINNER II A1 channel model for 10 m only is shown for the closed space (LOS and NLOS curves for 20 m are omitted since they are
identical to those of 10 m). The WINNER II channel model is too optimistic compared to real measurements. WINNER II NLOS curves are obtained by
taking the first arrival path with zero delay as the direct path between Tx and Rx since the model does not provide absolute delays.

1536-1276 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TWC.2019.2954318, IEEE
Transactions on Wireless Communications
7

on Table II, there is a positive correlation between ToA error TABLE III
and delay spread for 10 MHz, with ρ(τ̂ , τm ) > ρ(τ̂ , τrms ) E MPIRICAL MEAN µ AND STANDARD DEVIATION σ OF THE TOTAL
K- FACTOR KT AND T OA BIAS . PARAMETERS µ AND σ OF THE T OA BIAS
for all spaces and channel conditions. This is because signal ARE OBTAINED FROM LOG - NORMAL FITTING . I . E . NORMAL FITTING OF
with small bandwidth cannot resolve significant portion of log10 (τ̂100 (ns)). VALUES τ̂100 ≤ 0 HAVE BEEN DISCARDED FROM THE
the channel impulse response (CIR), which makes τ̂ more FITTING .

dependent on τm and somewhat on τrms . In the case of 100


MHz, LOS, closed and semi-open spaces this dependency is KT (LOS only) ToA bias (NLOS only)
Scenario
µ (dB) σ (dB) µ (ns-dB) σ (ns-dB)
lower since small portion of the CIR close to direct path is Closed 10 m −4.08 3.86 1.327 0.281
more influential on the ToA error compared to τm or τrms . space 20 m −4.09 5.52 1.670 0.353
It is interesting that correlations ρ(τ̂ , τm ) and ρ(τ̂ , τrms ) are Semi- 10 m −1.05 3.92 0.804 0.411
open 20 m 0.61 4.21 0.748 0.485
negative for 100 MHz, LOS and open space. Our assumption is space 30 m 4.46 3.82 1.204 0.440
that since values of τm and τrms are small and ≈ 1/B, which Open 10 m 3.88 2.89 1.274 0.469
indicates that most multipath components are in the proximity space 20 m 2.97 3.30 1.208 0.415
of direct path, direct path is better isolated with increased τm
and τrms , i.e. if some of the rays are moved further from the
direct path.
2) Lack of NLOS Bias in NLOS Channel: In NLOS
environment, direct path may be blocked and the first detec-
III. E XTENDED WINNER II M ODEL S UITABLE FOR table path delayed by τ1 causes the NLOS bias N LOSbias =
R ANGING τ1 − dGLOS /c, where dGLOS is the geometric distance bet-
A. Deficiencies of the WINNER II Model for Ranging ween Tx and Rx and c is the speed of light. In contrast to
communications, the random NLOS bias is important for the
Typical multipath radio channel is presented in Fig. 4. In
purpose of localization and needs to be modeled based on
general, it consists of the LOS component if there is no
real measurements. In [19] the WINNER II model for outdoor
obstacle between the transmitter and receiver, propagation
scenarios is extended by including the NLOS bias based on the
paths diffused by clusters marked with dashed lines and
geometrical information about the BS and MS locations and
specular or reflected paths represented with solid lines.
surrounding buildings heights and layout. In [12] the NLOS
bias statistics is obtained from measurements in outdoor-to-
indoor channel. The NLOS bias has been also estimated from
UWB measurements in [13].
3) Constant Delay Spread Over Distance: A major disad-
vantage of most, if not all, communication channel models in-
cluding WINNER II is that constant delay spread over distance
is assumed, i.e., delay spread is typically averaged over all
measurements. As we have seen from Fig. 2 and Table II mean
Tx Rx
and standard deviation of the τm and τrms vary noticeably over
distance as well as the type of environment. Several indoor
channel measurement campaigns that study such relationships
in LOS channels are [20]–[24]. Authors in [20] performed
Fig. 4. Wireless multipath radio channel rms delay spread measurements in LOS conditions inside
several rooms of different sizes and materials. The rms delay
Main deficiencies of the WINNER II model for ranging spread for all rooms increases with distance up to a maximum
applications are: √
value τrms (ns) = k f loor area, where k depends on the
1) Lack of Specular Components in LOS Channel: In reflectivity of the room walls. In a setup similar to the previous
a corridor environment with reflective walls strong specular work, [21] reports that the average rms delay spread increases
components may exist besides the LOS path and diffused with the room size but in contrast to the former study decreases
rays. These strong reflections increase the mean and rms with increasing separation. For a large exhibition hall in [22],
delay spread that could be even larger than in the NLOS rms delay spread first increases for about half of the hall and
case (Fig. 2b). Strong reflections are causing means of the then drops as the distance approaches the length of the hall. By
total K-factor KT to be ≈-4 dB (Table III) and negative utilizing the reverberation theory, mean and rms delay spread
values more than 80% of the time for closed space, which are modeled as a function of distance in [23]. Obtained rms
is comparable to values reported in [18] for similar corridor delay spread dependency on the distance is similar to that
environment. Other two building environments are more open in [20]. Extended work in [24] shows a good prediction of
areas with less prominent reflections for which the KT is delay spread versus distance for different room absorption
more positive. The KT is the power ratio of the direct coefficients and total room surface areas. Considering that
path over the.powers of reflective and
 multipath components, ToA error depends on delay spread according to Table II,
2 2 PL 2
KT = |α1 | |αR | + l=2 |αl | , where αR is gain of the this kind of work is very relevant for channel models used
reflected path. in localization.

1536-1276 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TWC.2019.2954318, IEEE
Transactions on Wireless Communications
8

TABLE IV
L IST OF DELAY RELATED PARAMETERS . U NIT OF PARAMETERS DSµ AND DRµ IS log10 ([s]), DSσ AND DRσ IS log10 ([ns]), Kµ , Kσ , KRµ AND KRσ
ARE EXPRESSED IN D B WHILE rτ HAS NO UNIT.

Scenario existing parameters (Table 4-5 in [8]) additional parameters


wLOS rτ DSµ DSσ Kµ Kσ DRµ DRσ KRµ KRσ - -
wN LOS rτ DSµ DSσ - - - - - - DBµ DBσ

B. Additional WINNER II Indoor Channel Parameters 2) NLOS: In the case of NLOS channel random NLOS
bias delay is added to the WINNER II generated CIR. The
Deficiencies of the WINNER II model discussed previously
NLOS bias is assumed to be log-normal distributed for the
can be overcome by introducing additional mechanisms that
same reasons as in [13], primarily due to simplicity. Modified
are in accordance with the field measurements. CIRs generated
NLOS CIR can be expressed as:
by the WINNER II model are modified as shown in Fig. 5.
Any other channel model, indoor or outdoor, can be extended
hN LOS (t) = hW (t − τB ), (9)
in the similar fashion.
where τB is a random delay with mean DBµ and standard
Generate and
hLOS
deviation DBσ , defined as
Add Random

Yes
Reflective
Component τB = 10(DBσ χ+DBµ ) , (10)
Channel
WINNER II
A1 Model
hW
LOS? + sinc
hOUT where χ ∼ N (0, 1).
Interpolation
The Table IV shows an overview of the existing WINNER
No Generate and
Apply Random
hNLOS II A1 [8] and additional parameters that are affecting mean
Delay Bias and rms delay spread. The existing parameters are delay
proportionality factor rτ , mean and standard deviation of the
Fig. 5. Flow diagram of the extended WINNER II model rms delay spread DSµ and DSσ , and mean and standard
deviation of Rician K-factor Kµ and Kσ . There are many
1) LOS: In the LOS condition only one additional path is other parameters that do not affect delay spread significantly
added for compensating possible specular components for the and they are irrelevant for this study. Additional parameters
sake of simplicity although more strong reflections may exist. KRµ , KRσ , DRµ , DRσ , DBµ and DBσ are implemented
The overall CIR can be expressed as: by modifying the WINNER II channel model toolbox available
on MathWork website [25].
hLOS (t) = hW (t) + αR δ(t − τR ) (4)
L
C. Joint Mean and RMS Delay Spread Fitting
X
= αl δ(t − τl ) + αR δ(t − τR ), (5) Since ToA error depends significantly on τm and τrms , es-
l=1 pecially for signals with lower bandwidths, determining delay
spread for specific indoor environment type and distance can
where hW (t) are CIRs of the WINNER II model, while αR ∈ lead to better prediction of the ToA errors through simulations.
C and τR ∈ R are random variables. Gain of the reflected path Our goal is to estimate parameters w ∈ {wLOS , wN LOS }
αR is modeled as in Table IV for all environments and distances by jointly
r
XL matching mean and rms delay spread statistics of the ex-
αR = KR |αl |2 · ej2πϕ , (6) tended model output hOU T to those obtained from the field
l=2
measurements. Number of channel realizations R = 500 are
ext ext
where ϕ ∼ U (0, 1) is random phase of the reflected ray and generated for a given w, τm and τrms are calculated and
meas meas
KR is reflective K-factor generated in the same fashion as the compared with τm and τrms resulting in Dτm meas ,τ ext and
m

regular Ricean K-factor K in [8] or Dτrms


meas ,τ ext , where Dn,m is the two-sample Kolmogorov-
rms
Smirnov (K-S) [26] goodness-of-fit test statistics defined as
KR = 10(KRσ ζ+KRµ )/10 , (7) Dn,m = sup |F1,n (x) − F2,m (x)|, (11)
x
where ζ ∼ N (0, 1), KRµ and KRσ are mean and standard de- where sup is the supremum function, and F1,n and F2,m are
viation of the reflective K-factor expressed in dB. Random va- the empirical cdfs of the first and second sample respectively.
riables K and KR are considered to be independent. Random An advantage of the K-S method is that it a distribution free
delay of the reflected path τR is log-normal distributed and test. Set of parameters ŵ that achieves the best statistical ma-
modeled as tch for both τm and τrms simultaneously is reached through an
τR = 10(DRσ ξ+DRµ ) , (8) optimization procedure that minimizes the following objective
function
where ξ ∼ N (0, 1), DRµ and DRσ are mean and standard

ŵ = arg min max Dτm meas ,τ ext , Dτ meas ,τ ext
m rms rms
. (12)
deviation of the reflected path delay. w

1536-1276 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TWC.2019.2954318, IEEE
Transactions on Wireless Communications
9

TABLE V
E XTENDED WINNER II PARAMETER VALUES ŵ OBTAINED FROM THE JOINT MEAN AND RMS DELAY SPREAD CDF FITTING

Parameters ŵ of the extended WINNER II model Goodness


Scenario existing parameters additional parameters of joint fit
rτ DSµ DSσ Kµ Kσ DRµ DRσ KRµ KRσ DBµ DBσ Dmax
Closed LOS 10 m 1.46 −6.97 1.0e−3 5.35 3.05 −7.00 1.1e−1 4.19 0.30 - - 0.09
space 20 m 1.77 −6.86 1.0e−3 5.17 3.20 −7.47 3.0e−1 3.06 1.12 - - 0.09
NLOS 10 m 2.28 −7.17 1.0e−3 - - - - - - −8.08 0.28 0.11
20 m 1.37 −7.00 1.0e−3 - - - - - - −7.61 0.36 0.18
Semi- LOS 10 m 1.03 −7.14 1.0e−3 2.07 2.06 −7.25 9.9e−3 0.01 3.39 - - 0.08
open 20 m 0.86 −7.15 1.5e−3 3.19 0.31 −7.77 5.3e−3 1.24 0.31 - - 0.15
space 30 m 0.78 −7.18 1.0e−3 5.03 2.20 −7.13 5.2e−2 −8.10 3.59 - - 0.08
NLOS 10 m 4.15 −7.30 1.0e−3 - - - - - - −8.93 0.40 0.07
20 m 3.47 −7.29 1.0e−3 - - - - - - −8.54 0.49 0.12
30 m 2.57 −7.06 1.0e−3 - - - - - - −7.92 0.43 0.14
Open LOS 10 m 2.32 −7.52 1.1e−3 5.47 0.26 −7.58 5.0e−3 1.80 0.35 - - 0.14
space 20 m 1.59 −7.33 1.0e−2 2.09 3.09 −7.81 5.2e−3 1.09 3.18 - - 0.11
NLOS 10 m 3.21 −7.41 1.0e−3 - - - - - - −8.09 0.50 0.15
20 m 3.47 −7.38 1.0e−1 - - - - - - −8.14 0.42 0.11

The Nelder-Mead simplex algorithm, which belongs to the NLOS conditions and distance. Path loss and shadowing are
class of derivative free optimizations [27], is used for finding included in the simulations per WINNER II specification [8].
a local minimum of the objective function. The algorithm is For fair comparison, each NLOS link on floor plans in Fig. 1
computationally simple and also efficient in terms of function is classified based on the number of walls between Tx and
evaluation numbers. It is one of the local search algorithms Rx as well as structure of those walls (light or heavy) and
for which selecting a good starting point w0 is important. additional path loss is applied accordingly. Number of channel
Initial parameters of w0LOS in (12) for delay proportionality realizations that correspond to the number of measurements
factor rτ is in the range between 1 and 4, DSµ are set to in Table I are generated and convolved with the transmitted
values log10 (µ(ns)) − 9, where µ(ns) are taken from the waveforms. The AWGN noise is added at the receiver to
column τrms in Table II. DSσ is kept to a low value of achieve signal SNR values similar to those from the real
≈ 10−3 to match measurement cdfs with steep slopes (low measurements. The received signals are correlated with the
variances). Good initial guess of Kµ , Kσ , KRµ , and KRσ transmitted waveforms and ToA estimates τ̂ are obtained from
can be found from the column KT in Table III and relations the β estimator (3). We also included the search and subtract
KT = K − log10 (KR + 1) and KTσ2 ≈ Kσ2 + KRσ2 . (SS) ToA estimator [29] to show validity of the extended
Strong reflections are usually coming from surrounding metal model. The SS algorithm can detect multipath components
objects so DRµ typically takes values between -8 and -7, in a non-separable channel. Simulated received signals are
or equivalently between 10 and 100 ns from the direct path. processed in the same way as the real measurement captures.
DRσ corresponds to the standard deviation of the delay of the
The ToA statistics from simulations are compared with
reflected path and takes relatively small values.
measured ToA cdfs on Figs. 6 and 7. In the case of β estimator,
Initial values of DBµ and DBσ in w0N LOS can be set to
ranging accuracy of the extended model is calculated from
µ(ns-dB) − 9 and σ(ns-dB) from the column ToA bias in
the absolute errors || at 10, 50 and 90% percentiles levels,
Table III. The use of unit τ [ns-dB] = log10 (τ [ns]) is necessary
with the mean absolute errors (MAE) of 2.4 and 1.6 m for
for comparison with results in [8], [28]. The rest of values are
10 and 100 MHz signals, and the root mean square errors
assigned as in the case of w0LOS .
(RMSE) of 3.3 and 2.7 m for 10 and 100 MHz signals,
Joint mean and rms delay spread fitting was done for all
respectively. Overall, the cdfs obtained from simulations show
environments, distances and propagation scenarios separately
good visual match to measurements with exceptions of closed
with number of optimization steps in (12) limited to 100.
 space, 20 m and NLOS (Figs. 6b and 7b), semi-open space,
Values Dmax = max Dτm meas ,τ ext , Dτ meas ,τ ext in Table V
m rms rms
20 m and NLOS (Fig. 6d), and semi-open space, 30 m and
are the final K-S test statistic values. The Table V provides
NLOS (Fig. 7e). This is not surprising since the corresponding
set of parameters for diverse environments but small number
Dmax values in Table V and || are among the largest. One
of distances. Although we provided a methodology for finding
anomaly from this observation is closed space, 10 m and LOS
ŵ, taking measurements at more Tx/Rx separations and deter-
(Fig. 7a) with very small Dmax = 0.09 but surprisingly high
mining parameters for any given distance is needed for better
|| = 40.2 ns for 90% level. This exception is elaborated in
applicability of the extended model.
the next section.
IV. S IMULATION R ESULTS
Ranging accuracy of the SS estimator is expressed in the
A. Comparison of ToA errors from Measurements and Simu- MAE of 2.0 and 1.8 m for 10 and 100 MHz signals, and
lations the RMSE of 2.9 and 5.0 m for 10 and 100 MHz signals,
Extracted parameters ŵ in Table V are used in simulations respectively. Biggest discrepancy between simulations and
for estimating the ToA errors for a particular building, LOS or measurements is found for closed space, 20 m and NLOS

1536-1276 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TWC.2019.2954318, IEEE
Transactions on Wireless Communications
10

(a) Closed space, 10 m (b) Closed space, 20 m

(c) Semi-open space, 10 m (d) Semi-open space, 20 m

(e) Semi-open space, 30 m

(f) Open space, 10 m (g) Open space, 20 m


Fig. 6. Measured and simulated ToA error statistics for 10 MHz

1536-1276 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TWC.2019.2954318, IEEE
Transactions on Wireless Communications
11

(a) Closed space, 10 m (b) Closed space, 20 m

(c) Semi-open space, 10 m (d) Semi-open space, 20 m

(e) Semi-open space, 30 m

(f) Open space, 10 m (g) Open space, 20 m


Fig. 7. Measured and simulated ToA error statistics for 100 MHz

1536-1276 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TWC.2019.2954318, IEEE
Transactions on Wireless Communications
12

(a) Delay spread cdfs (b) Correlator profile

(c) ToA error cdfs, 10 MHz (d) ToA error cdfs, 100 MHz
Fig. 8. Comparison between the WINNER II and extended model for the closed space, LOS and 10 m; mean and rms delay spread, normalized and averaged
correlator profile (top) and corresponding ToA error cdfs for 10 and 100 MHz (bottom).

(Fig. 7b). The cdf centers of the SS estimator in Figs. 6 and 7 LOS scenario in the closed space at 10 m distance is selected
are closer to 0 ns or true distance since the algorithm for due to severe reflections, for which the extended model is
finding optimum number of paths in [29] depends on the mean expected to outperform WINNER II as a result of added
of ToA error in addition to its variance. specular component. Although the extended model achieves
In general, the ToA cdf mismatch is smaller for 10 MHz better fit with Dmax = 0.09 compared to Dmax = 0.16 for
compared to 100 MHz signal. This is reasonable because the WINNER II, and better visual match of mean and rms delay
τm and τrms fitting is more relevant for 10 MHz signal since spread illustrated in Fig. 8a, it performs worse for 100 MHz
larger portion of the CIR falls into its “non-resolvable” region signal in terms of the ToA error cdf (Fig. 8d) while the 10 MHz
(∝1/B). In the case of 100 MHz signal, paths of the CIR that signal in Fig. 8c matches very well with the measurements.
are within ≈10 ns from the direct arrival have strong effect The reason is that the optimization algorithm finds a good
on the ToA error while the later ones are less influential [30]. delay spread match by tuning mean of the reflective path to
DRµ = −7 (Table V) or equivalent delay of 100 ns from the
direct path (Fig. 8b). This delay shift covers reflections at ≈65
B. Comparison Between the WINNER II and Extended model and ≈115 ns, which fall into “non-resolvable” region of the 10
MHz signal and affects the ToA estimate. For 100 MHz signal
We are also interested in how WINNER II and extended mo-
delay shift of 100 ns is outside of its “non-resolvable” region
del compare on the same set of measurement data. Parameter
and has very small effect on the ToA estimate while reflection
optimization for WINNER II is done in the same fashion as
closest to the direct path at ≈20 ns is left uncompensated. The
in all other previous fittings with the difference that WINNER
fitting process could be improved by ignoring CIR paths that
II model contains only existing parameters from Table IV.

1536-1276 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TWC.2019.2954318, IEEE
Transactions on Wireless Communications
13

(a) Delay spread cdfs (b) Correlator profile

(c) ToA error cdfs, 10 MHz (d) ToA error cdfs, 100 MHz
Fig. 9. Comparison between the WINNER II and extended model for the closed space, NLOS and 20 m; mean and rms delay spread, normalized and averaged
correlator profile (top) and corresponding ToA error cdfs for 10 and 100 MHz (bottom).

are outside the signals “non-resolvable” region. wide signals in 5.5 GHz ISM band. Delay of the direct path
NLOS scenario in the closed space and 20 m is chosen due between the transmitter and receiver is known precisely as
to the largest NLOS bias compared to all other measurements. a result of tight synchronization and accurately measured
Applied random NLOS bias controlled by parameters DBµ distance between the antennas. We demonstrate that WINNER
and DBσ allows shifting mean delay spread to the right II and similar channel models designed for communications
(Fig. 9a) and finding a better match for joint delay spread are not suitable for localization applications as they do not
fit of Dmax = 0.18 (Table V) compared to Dmax = 0.32 provide absolute delays in NLOS channels, assume constant
for WINNER II. This positive bias is clearly visible in the rms delay spread over distance and neglect strong specu-
correlator profile comparison in Fig. 9b, which is lacking lar components. An extended WINNER model that includes
in the case of WINNER II with recognizable exponentially one specular component and NLOS bias is proposed, while
decaying channel response starting from τ = 0. Improvement modeling delay spread as a function of distance is left for
in reduced gap between the ToA error for extended model and future work. Extended model parameters that match measured
measurements for both 10 and 100 MHz signal bandwidths in mean and rms delay spread statistics are derived from a
Figs. 9c and 9d is noticeable. fitting procedure, which are then used for simulating extended
channel model and comparing ToA error cdfs to those obtained
V. C ONCLUSIONS from the measurements. Good match between the two ToA
error statistics is demonstrated. In future work, we hope to
We executed an indoor measurement campaign to statis-
determine delay spread over distance relationship and evaluate
tically characterize the bias in ToA estimates incurred by
the extended model on system level simulations of different
multipath and NLOS reflections on 10 MHz and 100 MHz

1536-1276 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TWC.2019.2954318, IEEE
Transactions on Wireless Communications
14

TDoA localization algorithms that are applicable to LTE [18] C. A. Tan, A. R. Nix, and M. A. Beach, “Dynamic spatial-temporal
OTDOA and potentially future 5G solutions in the sub-6GHz propagation measurement and super-resolution channel characterisation
at 5.2 GHz in a corridor environment,” in Proceedings IEEE 56th
band. Vehicular Technology Conference, vol. 2, 2002, pp. 797–801 vol.2.
[19] W. Wang, T. Jost, A. Lehner, U.-C. Fiebig, and C. Gentner, “Physical-
Statistical Channel Model for Joint GNSS and Mobile Radio Based
ACKNOWLEDGMENT Positioning,” in Proceedings of the 25th International Technical Meeting
of The Satellite Division of the Institute of Navigation (ION GNSS 2012),
The authors would like to thank Fernando Perez-Cruz, 2012, pp. 953–959.
Chunhua Geng, Dmitry Chizhik, Sivarama Venkatesan, Mike [20] J. T. E. McDonnell, T. P. Spiller, and T. A. Wilkinson, “Characte-
Macdonald, John Pastalan, Mike Zierdt, Cuong Tran and rization of the spatial distribution of RMS delay spread in indoor
LOS wireless environments at 5.2 GHz,” in Ninth IEEE International
Dragan Samardzija for their helpful discussions, insights, and Symposium on Personal, Indoor and Mobile Radio Communications
support. (Cat. No.98TH8361), vol. 2, 1998, pp. 621–624.
[21] H. K. Chung and H. L. Bertoni, “Indoor propagation characteristics at
5.2GHz in home and office environments,” Journal of Communications
R EFERENCES and Networks, vol. 4, no. 3, pp. 176–188, 2002.
[22] S. Sczyslo, S. Dortmund, and I. Rolfes, “Determination of the delay
[1] S. Gezici, Z. Tian, G. B. Giannakis, H. Kobayashi, A. F. Molisch, spread of an indoor channel measurement campaign in the UHF band,”
H. V. Poor, and Z. Sahinoglu, “Localization via ultra-wideband radios: in Proceedings of the 2012 IEEE International Symposium on Antennas
a look at positioning aspects for future sensor networks,” IEEE Signal and Propagation, July 2012, pp. 1–2.
Processing Magazine, vol. 22, no. 4, pp. 70–84, 2005. [23] G. Steinböck, T. Pedersen, B. H. Fleury, W. Wang, and R. Raulefs,
[2] S. Adler, S. Schmitt, K. Wolter, and M. Kyas, “A survey of experimental “Distance Dependent Model for the Delay Power Spectrum of In-room
evaluation in indoor localization research,” in International Conference Radio Channels,” IEEE Transactions on Antennas and Propagation,
on Indoor Positioning and Indoor Navigation, 2015. vol. 61, no. 8, pp. 4327–4340, 2013.
[3] S. Fischer, “Observed Time Difference of Arrival (OTDOA) Positioning [24] ——, “Experimental Validation of the Reverberation Effect in Room
in 3GPP LTE,” Qualcomm Technologies, Tech. Rep., 2014. Electromagnetics,” IEEE Transactions on Antennas and Propagation,
[4] M. Lipinski, T. Wlostowski, J. Serrano, and P. Alvarez, “White rabbit: vol. 63, no. 5, pp. 2041–2053, 2015.
A PTP application for robust sub-nanosecond synchronization,” in [25] WINNER II Channel Model for Communications System Toolbox.
IEEE International Symposium on Precision Clock Synchronization for [Online]. Available: https://www.mathworks.com/matlabcentral/
Measurement, Control, and Communication (ISPCS), 2011, pp. 25–30. fileexchange/59690-winner-ii-channel-model-for-communications-
[5] R. Hach and A. Rommel, “Wireless synchronization in time difference system-toolbox
of arrival based real time locating systems,” in 9th Workshop on [26] Frank J. Massey Jr., “The Kolmogorov-Smirnov Test for Goodness of
Positioning, Navigation, and Communication, 2012. Fit,” Journal of the American Statistical Association, vol. 46, no. 253,
[6] F. Perez-Cruz, C.-K. Lin, and H. Huang, “BLADE: A Universal, Blind pp. 68–78, 1951.
Learning Algorithm for ToA Localization in NLOS Channels,” in 2016 [27] L. Rios and N. Sahinidis, “Derivative-free optimization: a review of
IEEE Globecom Workshops (GC Wkshps), 2016, pp. 1–7. algorithms and comparison of software implementations,” Journal of
[7] 3GPP, “Study on 3d channel model for LTE,” Tech. Rep. 3GPP 36.873 Global Optimization, vol. 56, no. 3, pp. 1247–1293, 2013.
(V12.2.0), 2015. [28] “WINNER II Channel models - Part II Radio Channel Measurement and
[8] “WINNER II Channel models - Part I Channel Models, D1.1.2 V1.2, Analysis Results, D1.1.2 V1.0, IST-4-027756,” Sep 2007.
IST-4-027756,” Feb 2008. [29] C. Falsi, D. Dardari, L. Mucchi, and M. Z. Win, “Time of Arrival
[9] V. Savic, J. Ferrer-Coll, P. Angskog, J. Chilo, P. Stenumgaard, and E. G. Estimation for UWB Localizers in Realistic Environments,” EURASIP
Larsson, “Measurement analysis and channel modeling for TOA-based Journal on Advances in Signal Processing, pp. 1–13, 2006.
ranging in tunnels,” IEEE Transactions on Wireless Communications, [30] B. Alavi and K. Pahlavan, “Studying the Effect of Bandwidth on Perfor-
vol. 14, no. 1, pp. 456–467, 2015. mance of UWB Positioning Systems,” in IEEE Wireless Communications
[10] S. Adler, S. Schmitt, Y. Yang, Y. Zhao, and M. Kyas, “Experimental eva- and Networking Conference, 2006. WCNC 2006., vol. 2, 2006, pp. 884–
luation of indoor localization algorithms,” in International Conference 889.
on Indoor Positioning and Indoor Navigation, 2014, pp. 291–299.
[11] W. Wang, T. Jost, C. Mensing, A. Dammann, and K. Fawaz, “Indoor
Propagation Effects on ToA Bias for Joint GNSS and Terrestrial Radio
Based Localization,” in VTC Spring 2009 - IEEE 69th Vehicular
Technology Conference, 2009, pp. 1–5. Zoran Latinović received the B.S. degree in elec-
[12] W. Wang, T. Jost, and U. Fiebig, “Characteristics of the NLoS Bias trical engineering from University of Belgrade, Bel-
for an Outdoor-to-Indoor Scenario at 2.45 GHz and 5.2 GHz,” IEEE grade, Serbia, in 1996, and the M.S. and Ph.D.
Antennas and Wireless Propagation Letters, vol. 10, pp. 1127–1130, degrees in electrical engineering from Polytechnic
2011. University and New York University, Brooklyn, New
[13] N. A. Alsindi, B. Alavi, and K. Pahlavan, “Measurement and Modeling York, in 2009 and 2019, respectively. He is cur-
of Ultrawideband TOA-Based Ranging in Indoor Multipath Environ- rently a Member of Technical Stuff at Nokia Mobile
ments,” IEEE Transactions on Vehicular Technology, vol. 58, no. 3, pp. Networks. He has previously worked at Iritel and
1046–1058, 2009. NetQuest Corporation. His current research interest
[14] Z. Latinović, C. Geng, and H. Huang, “Channel Measurements and includes wireless channel measurements and mo-
Performance of Indoor Time-of-Arrival Localization at 5GHz,” in 2018 deling for ToA localization.
IEEE Wireless Communications and Networking Conference (WCNC),
2018, pp. 1–6.
[15] Z. N. Low, J. H. Cheong, C. L. Law, W. T. Ng, and Y. J. Lee, “Pulse
detection algorithm for line-of-sight (LOS) UWB ranging applications,” Howard Huang received a B.S. in electrical engi-
IEEE Antennas and Wireless Propagation Letters, vol. 4, pp. 63–67, neering from Rice University in 1991 and a Ph.D.
2005. in electrical engineering from Princeton University
[16] E. S. Kim, J. I. Kim, I.-S. Kang, C. G. Park, and J. G. Lee, “Simulation in 1995. Since graduating, he has been a research
results of ranging performance in two-ray multipath model,” in 2008 engineer at Bell Labs where he has worked on
International Conference on Control, Automation and Systems, 2008, multiple antenna techniques and their application
pp. 734–737. in cellular networks. More recently, he is working
[17] J. A. del Peral-Rosado, J. A. López-Salcedo, G. Seco-Granados, F. Za- on technologies for enabling scalable and accurate
nier, and M. Crisci, “Evaluation of the LTE positioning capabilities under positioning of devices. Dr. Huang has taught as an
typical multipath channels,” in 2012 6th Advanced Satellite Multimedia adjunct professor at Columbia University and is a
Systems Conference (ASMS) and 12th Signal Processing for Space Fellow of the IEEE.
Communications Workshop (SPSC), 2012, pp. 139–146.

1536-1276 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

You might also like